We tried using HAProxy (to force connections to the closest datacenter) instead of relying on Hector to do the right thing and had problems with it. The client libraries we have used are smart enough to handle the node up/down case, and in most scenarios, if there is a slow node, they will route around the slow node.
On Fri, Jun 27, 2014 at 10:38 AM, Johan Edstrom <seij...@gmail.com> wrote: > I think you are going to be creating problems that the drivers were > designed > to avoid, it is not a good idea in general. > > /je > > On Jun 27, 2014, at 11:28 AM, Serge Fonville <serge.fonvi...@gmail.com> > wrote: > > > Hector is same way, if any node is slow to responds, times out or dies > hector will remove it from the pool leaving making it look like cluster > dead. The entire fault tolerant part of cassandra would be lost. > > Does this still apply if all nodes contain all data? > > > > Kind regards/met vriendelijke groet, > > > > Serge Fonville > > > > http://www.sergefonville.nl > > > > > > 2014-06-27 18:04 GMT+02:00 Chris Lohfink <clohf...@blackbirdit.com>: > > Hector is same way, if any node is slow to responds, times out or dies > hector will remove it from the pool leaving making it look like cluster > dead. The entire fault tolerant part of cassandra would be lost. > > > > Chris > > > > On Jun 27, 2014, at 11:00 AM, Michael Dykman <mdyk...@gmail.com> wrote: > > > > > NO, really it can't. I know little of Hector, but when using the > > > datastax driver, Cassandra provides a highly available set of > > > connections; a single Cassandra session will have explicit connections > > > to all of the node which make up a given cluster. > > > > > > The Cassandra server itself relies on a lot of cross-node talk which > > > means that they must be visible to each other. In this case, HA proxy > > > will just get in the way and make you less highly available. > > > > > > On Fri, Jun 27, 2014 at 11:10 AM, Richard Jennings > > > <richardjenni...@gmail.com> wrote: > > >> Can a Cassandra client such as Hector operate successfully behind a > HA Proxy > > >> where the cluster is represented by a single IP Address? > > >> > > >> Regards > > > > > > > > > > > > -- > > > - michael dykman > > > - mdyk...@gmail.com > > > > > > May the Source be with you. > > > > > >
